CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: A novel test for bias in decision-making Authors:  Camelia Simoiu - Stanford University (United States) [presenting]
Sharad Goel - Stanford University (United States)
Sam Corbett-Davies - Stanford University (United States)
Abstract: In the course of conducting traffic stops, officers have discretion to search motorists for drugs and other contraband. Scholars and criminal justice advocates have raised concerns that search decisions are prone to racial bias, but it has proven difficult to empirically evaluate these claims due to well-known limitations of current tests. We develop a novel statistical method for testing for discrimination. Namely, we use a hierarchical Bayesian latent variable model to infer latent race-specific thresholds of evidence that officers apply when deciding to search motorists. On a data set of six million police stops in North Carolina from 2009 to 2014, we find that the threshold for searching blacks and Hispanics is significantly lower than the threshold for searching whites, suggestive of racial discrimination in these interactions.